BdezuSZo/turbosens2

TurboSens2是TurboSens涡扇发动机退化基准的第二个场景，扩展了TurboSens1，引入双通道隐藏状态分离机械磨损和瞬态污垢，通过幻影观测映射使感知状态与真实机械状态可能分离，并添加发动机清洗操作。数据集包含六种区域原型（大陆、沿海、干旱、赤道、高原、工业），每种具有主导大气应力和退化原型先验，以及分层季节性天气过程和更长的片段（平均约1.4万次飞行）。数据以HDF5格式提供，包括传感器流、真实机械磨损状态、维护动作、访问类型、事件掩码、天气等信息，用于时间序列预测任务，支持世界模型的自监督预训练和探针评估。数据集为合成数据，不反映真实舰队统计，遵循CC-BY-4.0许可。TurboSens2 is the second scenario of the TurboSens turbofan degradation benchmark, extending TurboSens1 with a two-channel hidden state separating mechanical wear from transient fouling, a phantom observation map allowing sensed state divergence, an engine wash action, six region archetypes (Continental, Littoral, Arid, Equatorial, Highland, Industrial) each with dominant atmospheric stressors, a hierarchical seasonal weather process, and longer episodes (mean ~14k flights). The dataset is in HDF5 format, containing sensor streams, ground truth mechanical wear state, maintenance actions, visit types, event masks, weather data, etc., for time-series forecasting tasks, intended for self-supervised world model pretraining and probing evaluation. It is synthetic data and does not reflect real fleet statistics, licensed under CC-BY-4.0.